An equipment operation and maintenance data correlation analysis method based on a knowledge graph

Abstract

The application discloses a kind of equipment operation and maintenance data association analysis methods based on knowledge graph, it is related to equipment operation and maintenance technical field, including the following steps: step one: equipment operation data and maintenance record are collected, and operation and maintenance dataset is generated;Step two: execute time registration and entity normalization, and construct equipment operation and maintenance knowledge graph;Step three: generate candidate entity pair and establish restriction mapping;Step four: execute layer cluster adhesion, and generate adhesion state and adhesion hindrance state;Step five: construct probability state migration system and generate double analog business state;Step six: construct improved Fellegi-Sunter model and calculate association likelihood ratio;Step seven: generate candidate entity partition, partition joint likelihood and cut set conservation state;Step eight: screening target entity partition, merging equipment entity and outputting operation and maintenance data association result.The application improves the accuracy and reliability of equipment entity association.

Description

Technical Field

[0001] This invention relates to the field of equipment operation and maintenance technology, and in particular to a method for correlation analysis of equipment operation and maintenance data based on knowledge graphs. Background Technology

[0002] With the increasing demand for digital operation and intelligent maintenance of industrial equipment, technologies for unified correlation analysis of equipment operation data, maintenance records, and equipment topology have received widespread attention. Existing equipment operation and maintenance management systems mainly rely on equipment identifier matching, field similarity calculation, and manual rules for data merging. However, in practical applications, the following problems commonly exist: Equipment operation data and maintenance records come from different sources, and there is a lack of unified standards for sampling time, action time, and identification methods, which easily leads to time misalignment and entity duplication. This results in data for the same equipment being stored in a scattered manner, and data for different equipment being incorrectly merged. Existing knowledge graph construction methods usually determine entity relationships based on local attribute similarity, making it difficult to simultaneously use maintenance events, connection pointers, node levels, and time stamps to verify the consistency of local relationships. When faced with missing identifiers, name changes, and record conflicts, the stability of entity association results is low. Traditional probabilistic matching methods mainly perform weighted judgments on independent comparison items, which is difficult to describe the changes in causal structure before and after maintenance event state transitions and entity merging. This easily produces candidate results with high local matching probabilities but conflicting overall relationships, affecting the accuracy of the equipment maintenance knowledge graph and the reliability of maintenance data association results.

[0003] Therefore, how to provide a knowledge graph-based method for the correlation analysis of equipment operation and maintenance data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a knowledge graph-based method for analyzing the correlation of equipment operation and maintenance data. This invention combines layered cluster bonding, probabilistic dual-simulation refinement, and an improved Fellegi-Sunter model to achieve entity identification, relationship verification, and accurate merging of equipment operation and maintenance data. It has the advantages of high data correlation accuracy, strong result stability, and high reliability of knowledge graphs.

[0005] A knowledge graph-based method for equipment operation and maintenance data association analysis according to an embodiment of the present invention includes the following steps: Step 1: Collect equipment operation data and maintenance records, and generate an operation and maintenance dataset by combining the equipment topology; Step 2: Perform time registration and entity normalization on the operation and maintenance dataset, extract equipment entities, determine operation and maintenance events and relationships, and construct an equipment operation and maintenance knowledge graph; Step 3: Generate candidate entity pairs based on the equipment operation and maintenance knowledge graph, divide local relation domains, and establish constraint mappings; Step 4: Perform layer clustering on the local relation domain based on the constraint mapping to generate the glued state and the glue-blocking state; Step 5: Based on the operation and maintenance events, construct a probabilistic state transition system for the equipment entity, perform probabilistic dual-simulation refinement, and generate dual-simulation quotient states; Step 6: Construct an improved Fellegi-Sunter model, write the adhesive state, adhesive resistance state, and dual simulation quotient state into the comparison vector, and calculate the correlation likelihood ratio; Step 7: Generate candidate entity partitions and joint likelihood of partitions based on candidate entity pairs and correlation likelihood ratios. Construct merge hypothesis graphs and separate hypothesis graphs based on candidate entity partitions, extract necessary causal cut sets and generate cut set conservation states. Step 8: Delete candidate entity partitions that do not satisfy the cut set conservation state, select the target entity partition with the largest joint likelihood, merge equipment entities and output the operation and maintenance data association results.

[0006] Optionally, step one specifically includes: Collect equipment operation data with equipment identification and sampling time, aggregate equipment operation data according to equipment identification, and generate operation collection records; Obtain the operation and maintenance records containing the equipment identification and the time of the action; extract the operation and maintenance event type, operation and maintenance action content, and execution result from the operation and maintenance records; and generate operation and maintenance event records. Extract equipment nodes, node connection directions, and node levels from the equipment topology, and establish a topology mapping table between equipment nodes and operation acquisition records and maintenance event records based on equipment identifiers; The operation and maintenance event records are arranged according to the sampling time and action time, and the topology mapping table is written into the arrangement result to generate the operation and maintenance dataset.

[0007] Optionally, step two specifically includes: A unified time series is constructed based on the sampling time and action time in the operation and maintenance dataset. The operation collection records and operation and maintenance event records are mapped to the unified time series to generate time registration records. Based on the equipment identifier, equipment node, and node level in the topology mapping table, entity normalization is performed on the time registration records, and the time registration records mapped to the same equipment node are aggregated into equipment entity records. Extract equipment entities from equipment entity records, determine maintenance events based on maintenance event type, action time, and equipment entities, and determine maintenance relationships based on node connection direction, node level, and equipment entities associated with maintenance events. Equipment entities and maintenance events are registered as graph nodes, maintenance relationships are registered as graph edges, and time stamps are configured for graph nodes and graph edges according to a unified time series to construct an equipment maintenance knowledge graph.

[0008] Optionally, step three specifically includes: Based on the node hierarchy and operation and maintenance event type in the equipment operation and maintenance knowledge graph, equipment entities with the same node hierarchy and the same associated operation and maintenance event type are selected, and two different equipment entities are combined into candidate entity pairs. Centered on the equipment entities within the candidate entity pair, extract the operation and maintenance events and relationships connected to the equipment entities, organize the operation and maintenance events and relationships according to the connection direction of the operation and maintenance relationships, and generate a local relationship domain; Based on the operation and maintenance event type, connection pointer and time stamp, the candidate entity pairs are matched against the corresponding local relation domains. Operation and maintenance relationships with the same operation and maintenance event type, the same connection pointer and overlapping time stamps are identified as overlapping relationships. Map the operation and maintenance events and operation and maintenance relationships in the local relation domain to the overlapping relationship, record the correspondence between operation and maintenance events and the correspondence between operation and maintenance relationships, and establish a restricted mapping between local relation domains.

[0009] Optionally, step four specifically involves: Based on the constraint mapping, the local relation domains corresponding to the candidate entity pairs are mapped to the overlapping relations, and paired local relation sections are generated according to the correspondence between operation and maintenance events and the correspondence between operation and maintenance relations. The paired local relation sections are passed along the constraint mapping. The operation and maintenance event type and time stamp of the corresponding operation and maintenance event are compared item by item, and the connection pointer of the corresponding operation and maintenance relationship is compared. Paired local relation sections with consistent comparison results are registered as section consistent records, and paired local relation sections with at least one inconsistent comparison result are registered as section conflict records. Record the splicing cross-sections according to the connection direction and time mark, and register the splicing results that form a unique operation and maintenance event correspondence and a unique operation and maintenance relationship correspondence as the glued state; The local relation domain position corresponding to the conflict record of the local relation domain is recorded along the constraint mapping. Inconsistent operation and maintenance event types, timestamps and connection pointers are written into the local relation domain position to generate the adhesion obstacle state.

[0010] Optionally, the layer cluster bonding refers to mapping the local relation domains corresponding to candidate entity pairs to overlapping relations based on the constraint mapping, generating paired local relation cross-sections based on the operation and maintenance event correspondence and the operation and maintenance relation correspondence, comparing the operation and maintenance event type and time stamp of the corresponding operation and maintenance event along the constraint mapping, comparing the connection direction of the corresponding operation and maintenance relation, splicing the paired local relation cross-sections with consistent comparison results according to the connection direction and time stamp, registering the splicing result that forms a unique operation and maintenance event correspondence and a unique operation and maintenance relation correspondence as the bonding state, and registering the local relation domain positions corresponding to the operation and maintenance event type, time stamp and connection direction with inconsistent comparison results as the bonding obstacle state.

[0011] Optionally, step five specifically includes: Based on the operation and maintenance events associated with the internal equipment entities of the candidate entities, the operation and maintenance events are arranged according to the time stamp, each operation and maintenance event is registered as a state node, and the state nodes with consecutive time stamps are connected in chronological order to form a state transition edge, thereby generating a state transition record. The occurrence count of state transition edges and the total number of state transition edges derived from the starting state node are statistically analyzed in the state transition record. The ratio of the occurrence count of state transition edges to the total number of state transition edges is recorded as the state transition probability. A probabilistic state transition system is constructed based on the state nodes, state transition edges, and state transition probabilities. Based on the type of operation and maintenance event, the candidate entity pairs are assigned to the corresponding state nodes in the initial state partition. The probability quality of the state nodes transitioning to each initial state partition is calculated. State nodes with different probability quality are assigned to different state partitions. The probability quality is calculated and the state partitions are split repeatedly until the state partitions remain unchanged. The unchanged state partition is compressed into a dual-simulated quotient state. The affiliation relationship between the state node and the dual-simulated quotient state is recorded. The transition relationship between the dual-simulated quotient states is established based on the state transition edge and the state transition probability.

[0012] Optionally, step six specifically includes: In the Fellegi-Sunter model, matching and non-matching assumptions are set, the adhesive state, adhesive barrier state and dual simulation quotient state are set as comparison states, the matching conditional probability and non-matching conditional probability corresponding to the comparison states are set, and the probability iteration rule for updating the matching conditional probability and non-matching conditional probability based on the matching probability of candidate entities is set to construct an improved Fellegi-Sunter model. The candidate entity pairs are encoded into adhesive states as adhesive comparison terms, adhesive barriers as barriers as barriers comparison terms, and dual-simulated quotient states and the transition relationships between dual-simulated quotient states as quotient state comparison terms. A comparison vector is generated according to the order of adhesive comparison terms, barriers comparison terms, and quotient state comparison terms. The comparison vector is input into the improved Fellegi-Sunter model. The matching probability of the candidate entity pair is calculated based on the matching conditional probability and the non-matching conditional probability. The matching probability is used as the update weight under the matching assumption, and the difference between the unit value and the matching probability is used as the update weight under the non-matching assumption. The matching conditional probability and the non-matching conditional probability are updated. The matching probability calculation and conditional probability update are repeated until the conditional probability calculation results are the same in two consecutive rounds. The matching joint probability is generated based on the matching conditional probability, and the non-matching joint probability is generated based on the non-matching conditional probability. The ratio of the matching joint probability to the non-matching joint probability is determined as the association likelihood ratio, and the association likelihood ratio is written into the corresponding candidate entity pair.

[0013] Optionally, step seven specifically includes: Based on the correlation likelihood ratio, determine the merging and separation states of candidate entity pairs, combine the merging and separation states of candidate entity pairs, delete combinations with the same equipment entity partitioning results, and generate candidate entity partitions. Based on the aggregation of the correlation likelihood ratios of the merged and separated states contained in the candidate entity partitions, the joint likelihood of the partitions corresponding to the candidate entity partitions is generated. Compress the equipment entities in the merged state in the candidate entity partition into merge nodes, connect the operation and maintenance events and operation and maintenance relationships associated with the equipment entities to the merge nodes, and construct a merge hypothesis graph. Keep the equipment entities in the separated state independent and retain the associated operation and maintenance events and operation and maintenance relationships, and construct a separation hypothesis graph. The start and end operation and maintenance events are determined based on time stamps. The causal paths connecting the start and end operation and maintenance events are extracted from the merged hypothesis graph and the separate hypothesis graph, respectively. The minimum set of operation and maintenance relationships that can cut off all causal paths is registered as the necessary causal cut set. The necessary causal cut sets in the merged hypothesis graph and the separate hypothesis graph are compared according to the constraint mapping to generate the cut set conservation state.

[0014] Optionally, step eight specifically includes: Based on the cut set conservation state corresponding to the candidate entity partitions, delete the candidate entity partitions that do not satisfy the cut set conservation state, and retain the candidate entity partitions that satisfy the cut set conservation state. Read the joint likelihood of the reserved candidate entity partitions, compare the joint likelihoods of the partitions, and determine the candidate entity partition with the largest joint likelihood as the target entity partition. Merge equipment entities based on their merged status in the target entity partition, and keep equipment entities independent based on their separated status in the target entity partition. Assign the operation and maintenance events and relationships of the equipment entities corresponding to the merged status to the merged equipment entities. The equipment operation and maintenance knowledge graph is updated based on the merged equipment entities, the equipment entities that remain independent, operation and maintenance events, and operation and maintenance relationships, and the operation and maintenance data association results are output.

[0015] The beneficial effects of this invention are: This invention achieves time registration and entity normalization of equipment operation data and maintenance records, and constructs an equipment maintenance knowledge graph based on equipment topology. This enables data from different sources, with different time bases, and different identifier formats to be mapped to corresponding equipment entities. Based on local relation domains, restricted mapping, and layer cluster bonding, it verifies the consistency of maintenance event types, time stamps, and connection pointers. It can identify local relation conflicts and generate bonding obstacle states, reducing the risk of entity misassociation caused by missing equipment identifiers, name changes, and record misalignments, and improving the integrity and accuracy of equipment maintenance data organization.

[0016] This invention extracts the steady-state transition features of equipment entities through probabilistic dual-simulation refinement, and uses an improved Fellegi-Sunter model to fuse adhesive states, adhesive hindrance states, and dual-simulation quotient states, thereby improving the ability to characterize the true degree of association between equipment entities based on the correlation likelihood comparison. At the same time, it constructs a merge hypothesis graph and a separate hypothesis graph, and uses necessary causal cut sets and cut set conservation states to exclude candidate entity partitions with inconsistent causal structures. This can avoid erroneous merging caused by relying solely on local similarity, and make the output operation and maintenance data association results more reliable, stable, and interpretable. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a knowledge graph-based equipment operation and maintenance data association analysis method proposed in this invention; Figure 2 This is a schematic diagram of the layer cluster bonding process in the equipment operation and maintenance data association analysis method based on knowledge graph proposed in this invention. Figure 3 This is a schematic diagram illustrating the candidate entity partitioning and filtering of a knowledge graph-based equipment operation and maintenance data association analysis method proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A knowledge graph-based method for analyzing the correlation of equipment operation and maintenance data includes the following steps: Step 1: Collect equipment operation data and maintenance records, and generate an operation and maintenance dataset by combining the equipment topology; Step 2: Perform time registration and entity normalization on the operation and maintenance dataset, extract equipment entities, determine operation and maintenance events and relationships, and construct an equipment operation and maintenance knowledge graph; Step 3: Generate candidate entity pairs based on the equipment operation and maintenance knowledge graph, divide local relation domains, and establish constraint mappings; Step 4: Perform layer clustering on the local relation domain based on the constraint mapping to generate the glued state and the glue-blocking state; Step 5: Based on the operation and maintenance events, construct a probabilistic state transition system for the equipment entity, perform probabilistic dual-simulation refinement, and generate dual-simulation quotient states; Step 6: Construct an improved Fellegi-Sunter model, write the adhesive state, adhesive resistance state, and dual simulation quotient state into the comparison vector, and calculate the correlation likelihood ratio; Step 7: Generate candidate entity partitions and joint likelihood of partitions based on candidate entity pairs and correlation likelihood ratios. Construct merge hypothesis graphs and separate hypothesis graphs based on candidate entity partitions, extract necessary causal cut sets and generate cut set conservation states. Step 8: Delete candidate entity partitions that do not satisfy the cut set conservation state, select the target entity partition with the largest joint likelihood, merge equipment entities and output the operation and maintenance data association results.

[0020] In this embodiment, step one specifically includes: Collect equipment operation data generated during equipment operation, write the corresponding equipment identifier and sampling time for each piece of equipment operation data, group equipment operation data belonging to the same equipment entity into the same record set according to the equipment identifier, and arrange the equipment operation data in the record set according to the sampling time to generate operation collection records. Acquire maintenance records generated during equipment maintenance, write corresponding equipment identifiers and action timestamps for each maintenance record, parse the content representing event categories in the maintenance records to generate maintenance event types, parse the content representing operation processes to generate maintenance action content, parse the content representing operation completion status to generate execution results, and associate maintenance event types, maintenance action content, and execution results with equipment identifiers and action timestamps to generate maintenance event records; The process involves parsing the nodes representing physical objects in the equipment topology to generate equipment nodes, determining the node connection direction based on the connection start and end points in the equipment topology, determining the node hierarchy based on the subordinate position of the equipment node in the equipment topology, associating equipment identifiers with corresponding equipment nodes, and registering operation and collection records and maintenance event records to the corresponding equipment nodes to generate a topology mapping table. The operation and collection records are arranged according to the sampling time, and the operation and maintenance event records are arranged according to the action time. The operation and collection records and the operation and maintenance event records are assigned to the corresponding equipment nodes according to the equipment identifier. The node connection direction and node level in the topology mapping table are written into the arrangement result to generate the operation and maintenance dataset.

[0021] In this embodiment, step two specifically involves: Based on the operation and maintenance dataset, the sampling time in the operation and collection records and the action time in the operation and maintenance event records are read. The sampling time and action time are converted into a unified time base. The conversion results are arranged in chronological order to form a unified time series. The operation and collection records are written to the corresponding sampling time and the operation and maintenance event records are written to the corresponding action time to generate a time registration record. The equipment identifier corresponding to the time registration record is read from the topology mapping table. The equipment node is determined according to the equipment identifier. The subordinate position of the equipment node in the equipment topology is checked according to the node level. Time registration records with the same equipment identifier and the same equipment node are collected into the same record set. A unified equipment entity identifier is configured for the record set to generate the equipment entity record. Read the equipment entity identifier and equipment node from the equipment entity record, identify the physical object corresponding to the equipment entity identifier as the equipment entity, determine the event content according to the operation and maintenance event type, determine the event occurrence order according to the action time, associate the event content and event occurrence order with the equipment entity to form an operation and maintenance event, determine the connection direction between equipment entities according to the node connection direction, and determine the operation and maintenance relationship by combining the node level and the equipment entity associated with the operation and maintenance event. Equipment entities and maintenance events are registered as graph nodes, and maintenance relationships are registered as graph edges connecting graph nodes. Based on a unified time series, the time stamps corresponding to the sampling time are configured to the equipment entities, and the time stamps corresponding to the action time are configured to the maintenance events and maintenance relationships. The graph nodes and graph edges are organized according to the connection points of the equipment nodes to construct an equipment maintenance knowledge graph.

[0022] In this embodiment, step three specifically includes: Based on the equipment operation and maintenance knowledge graph, read the node level and associated operation and maintenance event type of the equipment entity, collect the equipment entities according to the node level, filter the equipment entities with the same associated operation and maintenance event type and different equipment entity identifiers within the same node level, and combine any two equipment entities that meet the filtering conditions according to the order of the entity identifiers to generate candidate entity pairs. Centered on the equipment entity within the candidate entity pair, read the operation and maintenance events and relationships directly connected to the equipment entity along the graph edge. Determine the connection direction based on the connection start and connection end points of the operation and maintenance relationship record. Arrange the operation and maintenance events and relationships according to the connection direction. Submit the arrangement results to the relationship range corresponding to the equipment entity to generate a local relationship domain. Read the operation and maintenance event type, connection pointer and time stamp in the corresponding local relation domain of the candidate entity pair, pair the operation and maintenance relationships with the same operation and maintenance event type and the same connection pointer in two local relation domains, compare the start and end positions of the time stamps corresponding to the paired operation and maintenance relationships, and determine the paired operation and maintenance relationships with overlapping time ranges as overlapping relationships; Using the overlapping relationship as the mapping benchmark, establish the operation and maintenance event correspondence relationship for the operation and maintenance events that form an overlapping relationship in the two local relationship domains, establish the operation and maintenance relationship correspondence relationship for the operation and maintenance relationships that form an overlapping relationship, record the mapping direction between the two local relationship domains according to the operation and maintenance event correspondence relationship and the operation and maintenance relationship correspondence relationship, and establish a restricted mapping.

[0023] In this embodiment, step four specifically includes: Based on the constraint mapping, read the two local relation domains corresponding to the candidate entity pairs, pair the operation and maintenance events in the local relation domains according to the operation and maintenance event correspondence, pair the operation and maintenance relations in the local relation domains according to the operation and maintenance relation correspondence, use the overlapping relationship as the pairing boundary to cut out the corresponding operation and maintenance events and operation and maintenance relations, arrange the cut-out results according to the connection direction, and generate paired local relation sections; The transmission order between paired local relation sections is determined along the constraint mapping. The operation and maintenance event type and timestamp of the corresponding operation and maintenance event are read in sequence, and the connection pointer of the corresponding operation and maintenance relationship is read. When the operation and maintenance event type is the same, the timestamp overlaps, and the connection pointer is the same, the paired local relation section is registered as a section consistency record. When there are inconsistencies in the operation and maintenance event type, timestamp, and connection pointer, the paired local relation section is registered as a section conflict record. Arrange the cross-section consistent records according to the connection direction, and connect the adjacent cross-section consistent records according to the order of the time stamp. Check whether each operation and maintenance event in the connection result corresponds to only one operation and maintenance event, and check whether each operation and maintenance relationship corresponds to only one operation and maintenance relationship. Register the splicing result that satisfies both check results as the glued state. Tracing back along the constraint mapping, the mapping position of the cross-sectional conflict record in the local relation domain is recorded. The positions corresponding to inconsistent operation and maintenance event types, non-overlapping time stamps, and inconsistent connection pointers are recorded respectively. The recorded positions are associated with the cross-sectional conflict records to generate the adhesive obstacle state.

[0024] In this embodiment, cluster bonding refers to mapping the local relation domains corresponding to candidate entity pairs to overlapping relations based on constraint mapping, generating paired local relation cross-sections based on the correspondence between operation and maintenance events and the correspondence between operation and maintenance relations, comparing the operation and maintenance event types and timestamps of the corresponding operation and maintenance events along the constraint mapping, comparing the connection pointers of the corresponding operation and maintenance relations, splicing the paired local relation cross-sections with consistent comparison results according to the connection pointers and timestamps, registering the splicing results that form unique operation and maintenance event correspondences and unique operation and maintenance correspondences as the bonding state, and registering the positions of the local relation domains corresponding to the operation and maintenance event types, timestamps, and connection pointers with inconsistent comparison results as the bonding obstacle state.

[0025] In this embodiment, step five specifically includes: Based on the operation and maintenance events associated with the candidate entity pair reading the equipment entity, the operation and maintenance events are arranged in order of time stamp. Each operation and maintenance event in the arrangement sequence is registered as a state node, and the operation and maintenance event type and time stamp are written into the corresponding state node. Adjacent state nodes in the arrangement sequence are connected in chronological order, with the connection direction from the state node with the earlier time stamp to the state node with the later time stamp. The connection result is registered as a state transition edge. The state nodes and state transition edges are collected to generate a state transition record. For a single state transition edge in the state transition record, count the number of times the starting state node transitions to the ending state node, and count the total number of state transition edges originating from the starting state node. Divide the number of transitions by the total number of state transition edges to obtain the state transition probability corresponding to a single state transition edge. Write the state transition probability into the corresponding state transition edge, and construct a probabilistic state transition system based on the state node, state transition edge, and state transition probability. Read the state nodes in the corresponding probabilistic state transition system for candidate entity pairs. According to the same division rule for operation and maintenance event types, classify the state nodes into the initial state partition. For a single state node, calculate the sum of the state transition probabilities of the state transition edges pointing to the target initial state partition. Use the sum of the state transition probabilities as the probability quality of the state node transitioning to the target initial state partition. Compare the probability quality of the state nodes in the same initial state partition. Classify the state nodes with different probability quality into different state partitions. Recalculate the probability quality based on the split state partitions and continue to perform state partition splitting until the state partitions remain unchanged. A single state partition that remains unchanged is compressed into a dual-simulation quotient state. The state nodes contained in the state partition are mapped to the corresponding dual-simulation quotient states. The affiliation relationship between the state nodes and the dual-simulation quotient states is recorded. The state transition edges and state transition probabilities from the state nodes in the starting state partition to the state nodes in the ending state partition are summarized. The summarized results are registered as the transition relationship between the dual-simulation quotient states.

[0026] In this embodiment, step six specifically includes: In the Fellegi-Sunter model, matching and non-matching hypotheses are established. The matching hypothesis means that the two equipment entities within a candidate entity point to the same equipment entity, and the non-matching hypothesis means that the two equipment entities within a candidate entity point to different equipment entities. The adhesive state, adhesive barrier state, and dual analog quotient state are registered as comparison states. Matching conditional probability and non-matching conditional probability are configured for each type of comparison state. The matching conditional probability represents the proportion of the comparison state under the matching hypothesis, and the non-matching conditional probability represents the proportion of the comparison state under the non-matching hypothesis. A probability iteration rule is set to re-statistically calculate the two types of conditional probabilities using the candidate entity's matching probability, and an improved Fellegi-Sunter model is constructed. Read the glue state corresponding to the candidate entity pair, and encode the result of forming a unique operation and maintenance event correspondence and a unique operation and maintenance relationship correspondence as glue comparison item. Read the glue obstruction state, and encode the obstruction result corresponding to inconsistent operation and maintenance event types, non-overlapping time stamps, and inconsistent connection points as obstruction comparison item. Read the dual simulation quotient state corresponding to the candidate entity pair and the transition relationship between the dual simulation quotient states, and encode the attribution relationship and transition relationship of the dual simulation quotient states as quotient state comparison item. Combine the encoding results according to the fixed arrangement order of glue comparison item, obstruction comparison item, and quotient state comparison item to generate the comparison vector corresponding to the candidate entity pair. The comparison vector is input into the improved Fellegi-Sunter model. The matching conditional probability and the non-matching conditional probability corresponding to each comparison item in the comparison vector are read respectively. The matching probability of the candidate entity pair is calculated based on the relative support relationship between the two types of conditional probabilities. The matching probability is used as the update weight of the corresponding candidate entity pair under the matching hypothesis. The difference between the unit value and the matching probability is used as the update weight of the corresponding candidate entity pair under the non-matching hypothesis. The weighted occurrence ratio of each comparison state under the matching hypothesis and the non-matching hypothesis is recalculated according to the two types of update weights. The matching conditional probability and the non-matching conditional probability are updated. The matching probability calculation and conditional probability update are repeated until the matching conditional probability and the non-matching conditional probability obtained in two consecutive rounds are the same. Read the matching condition probabilities corresponding to the same results in two consecutive rounds of calculation, combine the matching condition probabilities corresponding to each comparison item according to the order of arrangement in the comparison vector to generate the matching joint probability, read the corresponding non-matching condition probabilities, combine the non-matching condition probabilities corresponding to each comparison item according to the same order of arrangement to generate the non-matching joint probability, perform a ratio operation between the matching joint probability and the non-matching joint probability to obtain the association likelihood ratio corresponding to the candidate entity pair, and write the association likelihood ratio into the corresponding candidate entity pair.

[0027] In this embodiment, step seven specifically includes: Read the correlation likelihood ratios corresponding to candidate entity pairs. Determine the merging state based on the degree of support for two equipment entities belonging to the same equipment entity according to the correlation likelihood ratios. Determine the separation state based on the degree of support for two equipment entities remaining independent according to the correlation likelihood ratios. Combine the merging and separation states of candidate entity pairs in sequence. Gather equipment entities in the merging state according to the combination results and keep equipment entities in the separation state independent. Compare the equipment entity partitioning results, delete combinations with completely identical partitioning results, and generate candidate entity partitions. For a single candidate entity partition, read the candidate entity pairs in the merged and separated states, and aggregate the association likelihood ratios corresponding to the candidate entity pairs according to the merged and separated states. This allows the partition joint likelihood to simultaneously characterize the degree of matching support for the equipment entity merged relationship and the degree of non-matching support for the equipment entity separated relationship in the candidate entity partition, thereby generating the partition joint likelihood corresponding to the candidate entity partition. Read the merge status in the candidate entity partition, compress the equipment entities in the merge status into the same merge node, and transfer the operation and maintenance events and operation and maintenance relationships connecting each equipment entity before compression to the merge node. Construct a merge hypothesis graph based on the transferred operation and maintenance events and operation and maintenance relationships. Read the separation status in the candidate entity partition, keep the equipment entities in the separation status as independent nodes, and retain the original operation and maintenance events and operation and maintenance relationships of each equipment entity. Construct a separation hypothesis graph. Based on the time stamps corresponding to the operation and maintenance events, the operation and maintenance events in the merged hypothesis graph and the separated hypothesis graph are arranged. The operation and maintenance event corresponding to the start point of the time series is determined as the starting operation and maintenance event, and the operation and maintenance event corresponding to the end point of the time series is determined as the ending operation and maintenance event. The causal path connecting the starting operation and maintenance event and the ending operation and maintenance event is extracted along the connection direction of the operation and maintenance relationship. From the causal path, the minimum set of operation and maintenance relationships that can simultaneously cut off all causal paths and have no operation and maintenance relationships that can be deleted is selected and registered as the necessary causal cut set. Based on the constraint mapping, the operation and maintenance relationship correspondence of the necessary causal cut set in the two hypothesis graphs is established, and the cut set conservation state is generated according to the operation and maintenance relationship correspondence.

[0028] In this embodiment, step eight specifically includes: Based on the cut set conservation state corresponding to the candidate entity partition, read the necessary causal cut sets of the candidate entity partition in the merge hypothesis graph and read the necessary causal cut sets of the candidate entity partition in the separate hypothesis graph. Check whether the correspondence of operation and maintenance relationships in the two necessary causal cut sets is consistent according to the constraint mapping. The candidate entity partitions that can form a complete correspondence of operation and maintenance relationships are determined to satisfy the cut set conservation state. The candidate entity partitions that have interrupted correspondence of operation and maintenance relationships are determined to not satisfy the cut set conservation state. Delete the candidate entity partitions that do not satisfy the cut set conservation state and retain the candidate entity partitions that satisfy the cut set conservation state. Read the partition joint likelihood corresponding to the reserved candidate entity partitions, and use the partition joint likelihood as the evaluation result of the overall validity of the merged state and the separated state within the candidate entity partition. Compare the partition joint likelihood corresponding to the candidate entity partitions one by one, arrange the candidate entity partitions from largest to smallest according to the partition joint likelihood, determine the candidate entity partition with the largest partition joint likelihood corresponding to the first one in the ranking as the candidate entity partition with the largest partition joint likelihood, and register the candidate entity partition with the largest partition joint likelihood as the target entity partition. Read the corresponding merge and separation states of candidate entity pairs in the target entity partition, determine the equipment entities that need to be assigned to the same equipment entity according to the merge state, merge the equipment entities that need to be assigned to the same equipment entity, retain a unified equipment entity identifier for the merged equipment entity, assign the operation and maintenance events associated with the equipment entities before the merger to the merged equipment entity, assign the operation and maintenance relationships associated with the equipment entities before the merger to the merged equipment entity according to the connection pointer, keep the corresponding equipment entities independent according to the separation state, and retain the operation and maintenance events and operation and maintenance relationships associated with the independent equipment entities. The merged equipment entities in the equipment operation and maintenance knowledge graph are replaced with the merged equipment entities to maintain the independence of the equipment entities. The corresponding graph nodes in the equipment operation and maintenance knowledge graph are retained. The graph nodes are reconnected according to the relationship between operation and maintenance events and equipment entities. The equipment entities and operation and maintenance events are reconnected according to the connection points of the operation and maintenance relationships. This ensures that the equipment entity partitioning results in the equipment operation and maintenance knowledge graph are consistent with the target entity partitioning. The updated equipment entities, operation and maintenance events, operation and maintenance relationships and connection structures are used to generate operation and maintenance data association results and output them.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to an industrial equipment operation and maintenance data management platform. The platform accesses the operation data and maintenance records of power equipment, transmission equipment, pump sets, and auxiliary equipment, collecting a total of 28,640 equipment operation records, 4,360 maintenance records, and 1,280 equipment topology connection records, involving 620 equipment entities and 18 types of maintenance events. The original data contained issues such as inconsistent equipment name abbreviations, updated equipment identifiers, offsets between sampling and action times, duplicate registrations of the same maintenance event, and similar equipment names. Manual review revealed 86 duplicate equipment entity groups that needed merging, and 42 groups of similar-named but essential equipment entities. Traditional field matching methods easily merged "Circulating Pump No. 1" and "Circulating Pump Backup No. 1" incorrectly, and could also split the same equipment into two entities due to equipment identifier changes, affecting fault tracing and maintenance record queries.

[0030] When applying this invention, equipment operation data is first collected according to equipment identifiers. Operation data collection records are generated based on sampling times, and operation and maintenance event types, action content, and execution results are extracted from these records to generate operation and maintenance event records. Equipment nodes, node connection directions, and node levels in the equipment topology are used to establish a topology mapping table. Operation data collection records and operation and maintenance event records are registered according to a unified time series. For records where the sampling time and action time are offset, the equipment entity affiliation is determined by the equipment identifier, equipment node, and node level, forming an equipment entity record. After time registration and entity normalization, the platform constructs an equipment operation and maintenance knowledge graph containing 620 equipment entity nodes, 4360 operation and maintenance event nodes, and 7820 operation and maintenance relationship edges, configuring time stamps for graph nodes and edges.

[0031] In the equipment operation and maintenance knowledge graph, candidate entity pairs were selected based on node hierarchy and operation and maintenance event type, generating a total of 1540 candidate entity pairs. Local relation domains were divided around the equipment entities within each candidate entity pair, and overlapping relationships were determined based on operation and maintenance event type, connection direction, and time stamp, establishing a constraint mapping. Layered cluster bonding identified 1126 sets of cross-sectional consistent records and 414 sets of cross-sectional conflict records, including 176 sets of time stamp conflicts, 138 sets of connection direction conflicts, and 100 sets of operation and maintenance event type conflicts. A probabilistic state transition system was constructed based on operation and maintenance events, and probabilistic dual-simulation refinement was performed, compressing repetitive state nodes into dual-simulation quotient states. The improved Fellegi-Sunter model incorporated bonded states, bonded hindering states, and dual-simulation quotient states into the comparison vector, and calculated the association likelihood ratio of candidate entity pairs through conditional probability iteration, allowing local relation consistency, state transition consistency, and relation conflict information to jointly participate in equipment entity matching.

[0032] Candidate entity partitions and joint likelihoods of partitions are generated based on the correlation likelihood ratio, and merged hypothesis graphs and separate hypothesis graphs are constructed respectively. The platform extracts the causal path between the start and end maintenance events from the two types of hypothesis graphs, determines the necessary causal cutsets, and generates cutset conservation states based on the constraint mapping. Candidate entity partitions that do not satisfy the cutset conservation state are deleted, and the candidate entity partition with the highest joint likelihood is determined as the target entity partition. To verify the processing effect, a field matching scheme is set as the first comparison scheme, which judges similarity only based on equipment name, equipment identifier, and model; a basic Fellegi-Sunter scheme is set as the second comparison scheme, which calculates the matching probability based on attribute comparison items and does not use layer bundle gluing, probabilistic double simulation refinement, or cutset conservation states. As shown in Table 1: Table 1 Comparison of Equipment Entity Association Effects

[0033] Table 1 shows that the field matching scheme is affected by name abbreviation, identifier updates, and similar models, resulting in a high number of erroneous and omitted mergings. The basic Fellegi-Sunter scheme improves the entity association accuracy through probabilistic matching, but lacks local relationship bonding verification, state transition constraints, and causal structure verification, still resulting in the merging of entities with conflicting relationships. This invention writes the bonding state, bonding obstacle state, and dual analog quotient state into the comparison vector, and uses the cut set conservation state to filter out candidate entity partitions with inconsistent causal structures, achieving an entity association accuracy of 98.7% and an operation and maintenance relationship retention rate of 99.1%. Simultaneously, it reduces the number of erroneous and omitted mergings to two groups, demonstrating that this invention can improve the accuracy, stability, and processing efficiency of equipment operation and maintenance data association results.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based method for correlation analysis of equipment operation and maintenance data, characterized in that, Includes the following steps: Step 1: Collect equipment operation data and maintenance records, and generate an operation and maintenance dataset by combining the equipment topology; Step 2: Perform time registration and entity normalization on the operation and maintenance dataset, extract equipment entities, determine operation and maintenance events and relationships, and construct an equipment operation and maintenance knowledge graph; Step 3: Generate candidate entity pairs based on the equipment operation and maintenance knowledge graph, divide local relation domains, and establish constraint mappings; Step 4: Perform layer clustering on the local relation domain based on the constraint mapping to generate the glued state and the glue-blocking state; Step 5: Based on the operation and maintenance events, construct a probabilistic state transition system for the equipment entity, perform probabilistic dual-simulation refinement, and generate dual-simulation quotient states; Step 6: Construct an improved Fellegi-Sunter model, write the adhesive state, adhesive resistance state, and dual simulation quotient state into the comparison vector, and calculate the correlation likelihood ratio; Step 7: Generate candidate entity partitions and joint likelihood of partitions based on candidate entity pairs and correlation likelihood ratios. Construct merge hypothesis graphs and separate hypothesis graphs based on candidate entity partitions, extract necessary causal cut sets and generate cut set conservation states. Step 8: Delete candidate entity partitions that do not satisfy the cut set conservation state, select the target entity partition with the largest joint likelihood, merge equipment entities and output the operation and maintenance data association results.

2. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step one specifically involves: Collect equipment operation data with equipment identification and sampling time, aggregate equipment operation data according to equipment identification, and generate operation collection records; Obtain the operation and maintenance records containing the equipment identification and the time of the action; extract the operation and maintenance event type, operation and maintenance action content, and execution result from the operation and maintenance records; and generate operation and maintenance event records. Extract equipment nodes, node connection directions, and node levels from the equipment topology, and establish a topology mapping table between equipment nodes and operation acquisition records and maintenance event records based on equipment identifiers; The operation and maintenance event records are arranged according to the sampling time and action time, and the topology mapping table is written into the arrangement result to generate the operation and maintenance dataset.

3. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step two specifically involves: A unified time series is constructed based on the sampling time and action time in the operation and maintenance dataset. The operation collection records and operation and maintenance event records are mapped to the unified time series to generate time registration records. Based on the equipment identifier, equipment node, and node level in the topology mapping table, entity normalization is performed on the time registration records, and the time registration records mapped to the same equipment node are aggregated into equipment entity records. Extract equipment entities from equipment entity records, determine maintenance events based on maintenance event type, action time, and equipment entities, and determine maintenance relationships based on node connection direction, node level, and equipment entities associated with maintenance events. Equipment entities and maintenance events are registered as graph nodes, maintenance relationships are registered as graph edges, and time stamps are configured for graph nodes and graph edges according to a unified time series to construct an equipment maintenance knowledge graph.

4. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step three specifically involves: Based on the node hierarchy and operation and maintenance event type in the equipment operation and maintenance knowledge graph, equipment entities with the same node hierarchy and the same associated operation and maintenance event type are selected, and two different equipment entities are combined into candidate entity pairs. Centered on the equipment entities within the candidate entity pair, extract the operation and maintenance events and relationships connected to the equipment entities, organize the operation and maintenance events and relationships according to the connection direction of the operation and maintenance relationships, and generate a local relationship domain; Based on the operation and maintenance event type, connection pointer and time stamp, the candidate entity pairs are matched against the corresponding local relation domains. Operation and maintenance relationships with the same operation and maintenance event type, the same connection pointer and overlapping time stamps are identified as overlapping relationships. Map the operation and maintenance events and operation and maintenance relationships in the local relation domain to the overlapping relationship, record the correspondence between operation and maintenance events and the correspondence between operation and maintenance relationships, and establish a restricted mapping between local relation domains.

5. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step four specifically involves: Based on the constraint mapping, the local relation domains corresponding to the candidate entity pairs are mapped to the overlapping relations, and paired local relation sections are generated according to the correspondence between operation and maintenance events and the correspondence between operation and maintenance relations. The paired local relation sections are passed along the constraint mapping. The operation and maintenance event type and time stamp of the corresponding operation and maintenance event are compared item by item, and the connection pointer of the corresponding operation and maintenance relationship is compared. Paired local relation sections with consistent comparison results are registered as section consistent records, and paired local relation sections with at least one inconsistent comparison result are registered as section conflict records. Record the splicing cross-sections according to the connection direction and time mark, and register the splicing results that form a unique operation and maintenance event correspondence and a unique operation and maintenance relationship correspondence as the glued state; The local relation domain position corresponding to the conflict record of the local relation domain is recorded along the constraint mapping. Inconsistent operation and maintenance event types, timestamps and connection pointers are written into the local relation domain position to generate the adhesion obstacle state.

6. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, The layer cluster bonding refers to mapping the local relation domains corresponding to candidate entity pairs to overlapping relations based on the constraint mapping, generating paired local relation cross-sections based on the operation and maintenance event correspondence and the operation and maintenance relation correspondence, comparing the operation and maintenance event type and time stamp of the corresponding operation and maintenance event along the constraint mapping, comparing the connection pointer of the corresponding operation and maintenance relation, splicing the paired local relation cross-sections with consistent comparison results according to the connection pointer and time stamp, registering the splicing result that forms a unique operation and maintenance event correspondence and a unique operation and maintenance relation correspondence as the bonding state, and registering the position of the local relation domain corresponding to the operation and maintenance event type, time stamp and connection pointer with inconsistent comparison results as the bonding obstacle state.

7. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step five specifically involves: Based on the operation and maintenance events associated with the internal equipment entities of the candidate entities, the operation and maintenance events are arranged according to the time stamp, each operation and maintenance event is registered as a state node, and the state nodes with consecutive time stamps are connected in chronological order as state transition edges to generate state transition records. The occurrence count of state transition edges and the total number of state transition edges derived from the starting state node are statistically analyzed in the state transition record. The ratio of the occurrence count of state transition edges to the total number of state transition edges is recorded as the state transition probability. A probabilistic state transition system is constructed based on the state nodes, state transition edges, and state transition probabilities. Based on the type of operation and maintenance event, the candidate entity pairs are assigned to the corresponding state nodes in the initial state partition. The probability quality of the state nodes transitioning to each initial state partition is calculated. State nodes with different probability quality are assigned to different state partitions. The probability quality is calculated and the state partitions are split repeatedly until the state partitions remain unchanged. The unchanged state partition is compressed into a dual-simulated quotient state. The affiliation relationship between the state node and the dual-simulated quotient state is recorded. The transition relationship between the dual-simulated quotient states is established based on the state transition edge and the state transition probability.

8. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step six specifically involves: In the Fellegi-Sunter model, matching and non-matching assumptions are set, the adhesive state, adhesive barrier state, and dual simulation quotient state are set as comparison states, the matching conditional probability and non-matching conditional probability corresponding to the comparison states are set, and the probability iteration rule for updating the matching conditional probability and non-matching conditional probability based on the matching probability of candidate entities is set to construct an improved Fellegi-Sunter model. The candidate entity pairs are encoded into the corresponding bonding states as bonding comparison terms, the bonding hindering states as hindering comparison terms, and the dual-simulated quotient states and the transition relationships between the dual-simulated quotient states as quotient state comparison terms. A comparison vector is generated according to the order of bonding comparison terms, hindering comparison terms, and quotient state comparison terms. The comparison vector is input into the improved Fellegi-Sunter model. The matching probability of the candidate entity pair is calculated based on the matching conditional probability and the non-matching conditional probability. The matching probability is used as the update weight under the matching assumption, and the difference between the unit value and the matching probability is used as the update weight under the non-matching assumption. The matching conditional probability and the non-matching conditional probability are updated. The matching probability calculation and conditional probability update are repeated until the conditional probability calculation results are the same in two consecutive rounds. The matching joint probability is generated based on the matching conditional probability, and the non-matching joint probability is generated based on the non-matching conditional probability. The ratio of the matching joint probability to the non-matching joint probability is determined as the association likelihood ratio, and the association likelihood ratio is written into the corresponding candidate entity pair.

9. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step seven specifically involves: Based on the correlation likelihood ratio, determine the merging and separation states of candidate entity pairs, combine the merging and separation states of candidate entity pairs, delete combinations with the same equipment entity partitioning results, and generate candidate entity partitions. Based on the aggregation of the correlation likelihood ratios of the merged and separated states contained in the candidate entity partitions, the joint likelihood of the partitions corresponding to the candidate entity partitions is generated. Compress the equipment entities in the merged state in the candidate entity partition into merge nodes, connect the operation and maintenance events and operation and maintenance relationships associated with the equipment entities to the merge nodes, and construct a merge hypothesis graph. Keep the equipment entities in the separated state independent and retain the associated operation and maintenance events and operation and maintenance relationships, and construct a separation hypothesis graph. The start and end operation and maintenance events are determined based on time stamps. The causal paths connecting the start and end operation and maintenance events are extracted from the merged hypothesis graph and the separate hypothesis graph, respectively. The minimum set of operation and maintenance relationships that can cut off all causal paths is registered as the necessary causal cut set. The necessary causal cut sets in the merged hypothesis graph and the separate hypothesis graph are compared according to the constraint mapping to generate the cut set conservation state.

10. The method for correlation analysis of equipment operation and maintenance data based on knowledge graphs according to claim 1, characterized in that, Step eight specifically involves: Based on the cut set conservation state corresponding to the candidate entity partitions, delete the candidate entity partitions that do not satisfy the cut set conservation state, and retain the candidate entity partitions that satisfy the cut set conservation state. Read the joint likelihood of the reserved candidate entity partitions, compare the joint likelihoods of the partitions, and determine the candidate entity partition with the largest joint likelihood as the target entity partition. Merge equipment entities based on their merged status in the target entity partition, and keep equipment entities independent based on their separated status in the target entity partition. Assign the operation and maintenance events and relationships of the equipment entities corresponding to the merged status to the merged equipment entities. The equipment operation and maintenance knowledge graph is updated based on the merged equipment entities, the equipment entities that remain independent, operation and maintenance events, and operation and maintenance relationships, and the operation and maintenance data association results are output.

Images